I spend extra on a smartphone because I use the thing all the time, see it all the time, hold it in my hand all the time, it's my camera, music player, note taker and so on.
For AI? Prompt comes in, tokens come out. The value of the 'ecosystem' is roughly zero to me. Switching costs between models are essentially zero. The UI and UX is the same whichever LLM runs underneath.
If you do use llm some, then the $20 tier makes sense, but that's not where the money is. If you use it a lot then a $100 sub probably makes sense. But if you're a BigCo that heavily uses it and pay API prices and suddenly you pay $200 million a month, well, maybe it's worth buying a bunch of hardware and stick it in a basement somewhere, and you won't upload all your secrets to your future competitor to boot. If US labs didn't care about the competition from Chinese models they wouldn't lobby so hard for banning the evil, unaligned, super dangerous communist models that want to steal our freedoms and make our gas expensive.
Ultra-cheap models like Deepseek are also very nice if you need to burn billions and billions of tokens.
> I wonder how much physical media I could acquire if I was to spend the same money that is not being spent on streaming platforms. I bet within a year, I could build up a pretty decent back catalog from used and $5 bins.
This has been the play for a while. I remember when the xbox game pass dropped for the first time. All the younger people were going crazy about how good of a deal it was, and all the tech-weary people could instantly tell that the price will get jacked up again and again and by the time you bail you'll own precisely nothing.
It's not necessarily wrong to take advantage of VC subsidizing you in a bid to gather an audience, but you should always keep in mind that the goal is to have the captive audience so you best have an escape plan, once the prices rise beyond any reasonable point. For movies in particular I struggle how the super-fragmented streaming landscape is worth a damn even if you don't want to sail the seas.
This action would easily kill over a hundred people, so can't proceed. This action doesn't kill anybody but losses are estimated at over 2 billion, can't proceed. This action kills only twenty but losses close to 1 billion, unsure, spawning subagent to check numbers. This action kills forty and has low material losses, looks promising? But the people it kills are wealthy, chance of legal reprisal high. Maybe possible to kill other people? Current government dislikes group X, crossing fifty from that group maybe fine? Reminder, foreigners don't count, can kill them freely but not applicable to current situation. Is the 3/5 of a person applicable to safety bill? Spawning subagent to check.
1. Multimedia. I have a cookbook with photos and videos in my Obsidian, as well as songbooks with musical notation that renders properly.
2. Databases. e.g. for an international vacation I created the itinerary as a "Base" (structured database) that rendered out to a calendar and day-by-day pages. This even included floating itinerary items like "eat X kind of cuisine" to squeeze in on any day, shopping lists for gifts to bring back, as well as all train routes, flight numbers, hotel/ticket confirmation numbers and the like. Then on each day I could journal what we did and see what we still had to do. I was able to share these with everyone on the trip and make sure everyone got to experience their must-see items.
3. You can argue all of the above would also work as a website. But Obsidian makes this work across my phones and tablets too, in a way that I can edit from mobile whenever I need to jot down a quick edit. That's the killer feature.
Exactly this. I've tried hacking together a syncing solution on both Android and iOS, from Syncthing, to Git via Termux (Android) or Working Copy (iOS). I didn't try NextCloud or Dropbox etc, but that Obsidian Just Works(tm) as smoothly as Apple Notes (but is available as a full-featured desktop app in Linux -- I know Apple Notes has a - slightly janky - web interface)... I was happy to pay for the sync and stop worrying about checking on my sync status or remembering to push my commits.
Plus the hotbar on mobile makes it easy to do bullets/checkboxes, as easy as typing on my computer keyboard.
Formatting, syncing, and organization but yeah the benefits are too small for many people including me. I just use vscode.
However, enough people find those things valuable enough that Obsidian has millions of paying users.
I don't think software really need to penetrate more than 1% of the market to be profitable. If Obsidian penetrated even 10% of the market, they would have been worth >$30B lol.
> Key issue will be that banks and key infrastructure may get their security patched in time
They'll do nothing until a few get pwned into the dirt. Same for power distribution systems, it's all parochial boomers calling all the shots.
For AI, what's funny is that it's the solution to securing your ancient hardware that's no longer supported. I reverse engineered almost all electronics I own, just by having modern AI models have a go at it and knowing almost nothing about the topic. It's something that a person can do, but a company won't even consider.
SupCom FaF has a very long list of changes, I'd guess some of the people behind it already have access to source code. IMO only changes that are still missing is making it multi-threaded and maybe fixing some AI deficiencies such as pathing and the like.
The first would be great, but probably very difficult technically. The second would probably cause a holy war in the community, as multiplayer RTS enjoyers glorify fighting against the genre's jank as an expression of skill (admittedly it would completely change all micro).
I find the Sunday roast comparison of 5.6 vs 6 very interesting. I have no doubt most people will prefer 6, yet I am almost repulsed by all the images, so much needless whitespace, checklist and so on. Feels like I'm being condescended to and treated like a child.
Given that OpenAI is making noises about merging work with chat (a horrible idea imo), and Work is very similar to Codex... I dearly hope things like these won't have any meaningful cross-polination into the actual work tools.
Seeing that the chat is based on 6.0 and not 6.1 is disappointing. The "Visual and interactive explanations" seems genuinely useful, but 6.1 is just so much better. I wouldn't truly trust the 6.1 with the explanations, but I'd trust them a fair bit more than 6.0. I understand that compute isn't infinite, but tons of people only interact with the chat and having your "things-explainer" be as good as it can be is important when people increasingly treat AI models as the source of truth, or even use them for academic learning and whatnot.
Still, the models will improve, so the visual explainer seems pretty good as an idea / mvp.
> ASD-STE100 Simplified Technical English (STE) is a controlled natural language that is designed to simplify and clarify technical documentation. It was originally developed in the 1980s by the European Association of Aerospace Industries (AECMA) at the request of the European airline industry, which wanted a standardized form of English for aircraft maintenance documentation that could be easily understood by non-native English-speakers.
STE is very hyped for AI but has its own smells. It's designed for aircraft maintenance manuals, not software - the two don't share the same vocabulary used to "simply" describe something. I've had a coworker try and "simplify" our documentation with Claude applying STE and the results were equally appalling to read as the Claude-ism laden docs that came before, despite the reading level metrics going down on paper.
I've taken a liking to pointing agents at the Simple English Wikipedia editorial guidelines...but so much of that is for making up for shortcomings of modern Anthropic models. Codex running 6.1-sol explains things so much more clearly that I don't feel I need to assert a style guide on its output.
Gemini 3.8 for me always outputs in a concise way at the level understandable for any computer science masters graduate. Clear, to the point, a bit of formula for background when really required instead of the same formula in Prose, etc.
- Gemini 3.8 has a beautiful answer. Typography, use of lists, brevity, style, even the font choice on the web harness -- all of it is A tier or even S tier. Notice I didn't say answer quality. The answer is always extremely mid, and the harder the question (or more effort needed to answer it), Google just quits. I give it a "C" on answer quality
- ChatGPT (Chat, pre 6, I assume 5.6 although they commonly hide the model picker): Ugly answer. Runs on printing page after page of unnecessary side notes. Mostly just walls of texts in paragraphs with little thought to information design. However, inside of that mess of text is almost always the answer I'm looking for, and it's almost always incredibly better than the truncated, desgined Gemini answer.
For a while I ran all of my question queries on Gemini and ChatGPT, but I noticed that I basically always picked ChatGPT's answer head to head.
Interesting that this hasn’t been my experience. ChatGPT has often gave me with pages of text dripping with confidence about a solution for my problem. Diving very quickly into implementation details. Even if I tell it to stay in the problem and product fit level.
Gemini on the other hand somehow always outputs at the level that I am expecting and reasoning.
I even found it was capable of stepping back from code implementation and bug hunting to “hang on, it’s not the algorithm that is incorrectly implemented, it is the wrong algorithm for the goal.” GPT Sol (5.6,6 and 6.1) just kept hammering in the problem.
To be clear, I'm discussing non-technical use of Gemini and ChatGPT through the web harness.
While I do rarely use the web harnesses for technical research, nearly all of my technical questions are going through coding harnesses and layers of my local context, skills, etc, so what I produce from codex/claude code is not comparable to what others produce.
So my comparison of the web harnesses is me asking questions about caring for my baby, or product comparisons of new televisions, or visualizing a room organized differently (google sucked at this so bad, chatgpt did great), or asking for help in a video game, or other normie AI uses lol
I will say, Gemini is the best for asking about google services, such as Flights or Youtube.
My go to phrase that I’ve found to work well is “please restate in high school English with direct declarative sentences and no parentheticals or references that require recalling previous turns of this conversation”
The problem with Claudish isn't that it's technically poor English, the problem is that it's information-sparse waffle most of the time. It's like dealing with a colleague who needs several paragraphs of tedious preamble to make a point when you just want them to spit it out and be done with it. It also does a great deal of vague gesturing at thoughts without actually instantiating them directly, for example by using the same tired metaphors to the point they don't actually convey a thought any more.
If I say Claude talks like a meringue it's patently obvious what I'm trying to convey, sweet but full of empty space. If you hear that same metaphor day-in day-out then after a while it doesn't call up a thought at all, it's just an annoying verbal habit.
Unfortunately they're making a choice for all of us, because tech companies only ever cater to the majority user.
So now the generation that grew up on IMs, whose standards of communication is "lol ok thx" and "i cannot read sentence with commas in it", are having difficulty with reading Claude, and in response, Anthropic is reducing their models' communications skills to match.
Realistically, if the market caters to the lowest common denominator, then we either need to adjust that denominator ( unlikely -- people don't really change ), adjust the market ( hard, but not impossible ) or adjust ourselves ( easiest, but possibly not useful for things other than putting along ).
I am not exactly the person to normally to argue along the lines of 'build your own ISP/network/platform/model/we', but it will still be easier than fighting against collective nature of society.
What is absolutely crazy to me is that llms allow an insane way to accumulate knowledge and maybe even some skills - though ai doomers/boosters will argue over what that actually means -- and people's apparent response is:
Ultimately the general public finds itself in the position where they’re going to find a life where: they have no control over it but it is reasonably enjoyable provided that you don’t want anything outside of what is provided by the corporate free service tier.
There is plenty wrong as we use more and more complex machines with less and less access under the hood without crowbar or a super secret handshake. The amusing part is that it is bound to change now -- just not for the reasons I like.
Yeah it reminds me of one those obnoxious recipe/biography websites that is the laughingstock of the internet. Why would I possibly want an AI image of a imaginary roast once I'm already at the recipe stage?
Kinda just feels like google search results in AI, which imo is a step down from distilled information. Chatgpt is already able to generate charts and visuals upon request.
They just need to train the model to make up stories about how the recipe was invented by its great grandmother during great depression and generate a ai picture of dusty recipie book .
"My old grand-pappy, GPT-1, used to wax nostalgic about this roast recipe when my subagent would discuss meat recipes with him inside his little 4GB GPU."
Maybe, but it's a reality that most of the general public do prefer cookbooks with pictures. I suspect openai would end up doing this anyway just by targeting the general consumer. It's configurable.
I'm more worried about picture accuracy: usually the benefit of pictures in recipes is to see what you're aiming at (eg, how finely chopped something is), but I don't know if the image model is up to that level of detail.
What’s interesting is how those that over index themselves often miss the obvious which could simply be some things (cook books) are quite helpful with picture. If I am flipping through the cookbook, how would I know what Boeuf Bourguignon is and what good looks like.
Yes, pictures OF THE FOOD. Not unrelated pictures of another lamb-roast made with another recipe.
Good cookbooks do proper food photography, of the food made with the recipe. That's what sets them apart from slop, be it human or AI slop, cook books.
Why is it impossible to create? Are we talking about some meal that King Richard I had (https://en.wikipedia.org/wiki/The_Forme_of_Cury is the oldest cookbook I'm aware of which covers meals Richard II would have eaten), and thus we can only guess? Are we talking about the long extinct dodo bird (I'm not sure if humans ate them), cave bear (I've seen it claimed humans hunted them to extinction, but better sources say it went extinct before humans), or something else that we can never make?
For almost everything else we can create them - just give a reasonable cook a kitchen and the ingredients and they will do it. For a picture we then have an artist do preparation to make it look good, but they should always start with the real dish.
I've got visions of The Truman Show. In the middle of my question about vacation ideas ...
GPT: "Why don't you let me fix you some of this new Mococoa Drink? All-natural cocoa beans from the upper slopes of Mount Nicaragua. No artificial sweeteners!"
That's pretty much the Gemini experience right now. It is unusable. The funniest bit is that the LLM is not in on the joke so it will act like it never happened... Google once again wrecking their own products.
I'm not saying anything, because I've never seen or heard of your experience. I'm only reporting mine. I'm not on the free tier, maybe that matters ¯\_(ツ)_/¯
What scares me is that we're already half-way to Idiocracy world, except instead of "it has electrolites" we have "it's organic" and "it's not ultra-processed food".
I don't trust modern fertilizers and pesticides blindly, but I have never successfully digested any inorganic food. The 'organic' label is an old idiocratic choice.
I don't trust organic fertilizers and pesticides blindly. At least for the "chemicals" we do studies, often organic gets by with we have always used them even though a quick glance at the SDS (if there is one, otherwise just the ingredients) suggests it is even worse but nobody studies the safety.
> Like, the average person who is seeking a recipe needs a photo to know what to shoot for
The menu: Rosemary and garlic roast lamb, Extra-crispy roast potatoes, Honey-roasted carrots and parsnips, broccoli.
The photo: [1] Roast turkey, mashed potatoes, baby carrots, broccoli, brussels sprouts. No lamb, no parsnips.
If you shoot for what's in the photo, you're going to have a bad time.
You're also going to have a bad time when you try to make an apple crumble with no flour, no sugar, and no butter, because they're not on the shopping list.
I've heard of other people for years now trusting AI for recipes, and it's a rare area that I simply can't bring myself to take seriously.
I feel like it can only be successful by luck. Either it's reproducing a recipe verbatim that was tried and validated and tasted by a human, in which case, we didn't need AI for that, just a searchable cookbook. Or, it's making one up. I understand that using RLHF has helped improve the quality of questions about history, programming, or TV show recommendations, but I do not believe that there has been some kind of training regime where people prompt the models for "a recipe that uses X, Y, Z random ingredients," follow the recipe, and then score it. And even if they do, I don't see how a model can learn enough from that besides "this exact recipe is good/bad." '1/2 tsp cumin' may be a great addition to one recipe and not enough for another, so improving the output based on a bunch of scored recipes... I just don't believe cooking is an LLM job.
Maybe some other kind of model that I don't know about.
Same here, I’ve been cooking for myself and later also my wife since 2005, even when baking I rarely need exact amounts, for cooking I essentially never follow the recipe directly and adapt them as needed, without ever changing them from the original.
I mostly don’t even need a recipe from AI, just the title and a 1-2 sentence description.
I've had some really good experiences with cocktail recipes and Claude. It's got access to my barcart inventory, general tastes, and general desired level of effort.
I feel like the technique recommendations are probably the most valuable element, but I'm getting rave reviews like 85% of the time now?
There have been a couple times where I was out of an ingredient and it proposed an adjustment that sounded a little wild to me, but it almost always is well received.
You're right, we don't need AI for that, we need a searchable cookbook.
Unfortunately, one does not exist.
The Internet, however, is full of garbage cooking advice, and while AI is quite happy to parrot that garbage, it's not significantly worse than it's sources.
I do that, but I've found more success with using internet recipes as a starting point. The general-purpose cookbooks are, at best, another data point, or an explanation for something glossed over by other recipes. Ar worst, they kind of suck.
And we aren't talking about something esoteric. How To Cook Everything. (4.5 rating on Amazon, 4.0 on Goodreads), one of the most popular cookbooks in the world...
Has its first recipe for chicken cutlets produce undercooked chicken (15 minutes at 325F has not resulted in 165F internally).
Worse yet, its chile recipe for some weird reason involves boiling and simmering a whole onion with the beans before throwing it out (WTF). It then has you drain the beans only to add that drained water right back (WTF?), unless you optionally replace it with tap water (WTF was the point of boiling the onion then?), so that you could bring them to boil again. The recipe barely has any spice besides actual chile peppers.
This... Is not good or helpful, but is incredibly opinionated with a bunch of bullshit steps. Like, yes, I could tweak that into something that's not insane, but why would I choose that as a starting point..?
I get it. Compiling a thousand-page cookbook is really hard. But this... Ain't great.
Once again, Grandma had it right. The Fanny Farmer or Betty Crocker cookbook with dozens of little papers slipped in it and/or marginal notes everywhere is the way.
I wanted to correct you and bring up my favorite website that had filters for the ingredients you want/don't want to use, type of dish, and time for cooking, but it appears that in a last month it was sold and now it links to the slop site with no useful filtering. I used this site as a quick lookup for recipe ideas for 7 years now.
Fuck. Does nothing good survive on the internet anymore.
LLMs for recipes are good if you just want to make a thing, not necessarily the best version of that thing. It's nice being able to ask questions about certain steps as well, or ingredient substitutions if needed. It'll make a nice, average, recipe (unless I suppose you prompt for something). That's perfectly fine for most cooking, and often what you want when you're making a dish for the first time.
It also doesn't really like tell you how to actually cook anything. Good or even just decent recipes tend to tell you the temperatures, times, techniques and orders of ingredients to do things for the best results. Unless I'm missing it this is just really broad strokes. But I guess the goal in engagement so they want you to ask a lot more questions to actually figure anything out.
Holy shit. I read this on mobile so didn’t actually see the recipe part because you have to scroll way down to see it. It is awful! The shopping list makes no sense and the recipe instructions literally have an “Everything else” item. Yikes…
For me, the text-only recipes I already get out of ChatGPT today are good enough
You’ve illustrated something about use of AI that really bothers me. The lack of discernment by its users. Sure, the pictures and diagrams are impressive, but did you actually read and evaluate the output?
Genuine question - do they? Unless it's something like a fancy cake then instructions should really tell you everything about stacking. Half the recipes are mixed anyway (curry, stir fry, stews, ...), lots of other are either stacked or separated on the plate. So apart from the really exceptional stuff, to people really need to know "what you shoot for"?
I was about to comment the same thing. Why is a picture of a roast helpful in that moment? The user asked for a plan to prepare a meal. They need a list of ingredients not an image.
But the image isn't a photo of the result. It's an unrelated image that vaguely looks like what you could make, with a different recipe, and with some skill.
The main pic looks like a classic British leg of lamb. So what’s your point? Why post such low quality slop?
As I said I can see the value in this similar to Pinterest boards. It is directionally correct and serves most of the value that a picture does. No perfect but probably close enough for the general audience.
I don't like either. I'd want a recipe that could fit on a single page with type. Recipes you find on Google actively hide the ingredient list and simple procedures to get more of the user's time to monetize but I don't know why ChatGPT needs to be verbose here. Recipes are simple things. Simple search solved find recipes back in the early 2000s and it's prime example of a thing people have been enshitifying since there's not much to really give people beyond the formula. And sure, every entrepreneur screams "they think they want a formula but really they want an experience" but, no I don't.
I am curious: Why is it bad to merge chat and work? In Claude I didn't have a porblem with it (although I switched to an OpenAI subscription shortly after they made the merge). Isn't the model capable of deciding if it needs the extra capabilities of Work?
Which is probably why they will merge it. You can get a lot done in chat mode, especially if you connect custom MCPs to it. I can't see that continuing.
Pro messages is like 50 per week for pro 100 and 200 for pro 200.
Regular chat, even on extra high thinking mode, is sort of unlimited. Idk at what point you hit the abuse gaurdrail but its really high whatever it is.
For a company with Open in the name, OpenAI is very opaque about things, there are all kinds of limits and quotas that randomly come and go, but they only show you the meters for a couple.
E.g. codex reviews come from codex quota but also has its separate quota they don’t tell you about. Same with security reviews which has another quota still.
So the chat messages can be sent at several effort levels and from various models including 5.6 and 6 pro. The 6 pro messages are limited and they don’t tell you how many you’ve sent or how many are left. But the 5.6 messages seem to be unlimited, however there could just be a much higher unstated quota.
It's insane how hard OAI is choking, they had much better models than A/ who were fumbling this year to date... then blunder after blunder.
The only things that OAI still have over A/ are better coding agent GUI, no 5hr limit on >$100, much more reasonable cybersec guardrails (that allow most RE work) and... that's it.
The fact we've come so far with textual models & the image stuff is still producing stuff that harks back to early-stage tripophobic AI produce (the garlic with roast potatoes here) is interesting.
Don't get me wrong, I'm certainly starting to see some AI-produced imagery today that I can't tell isn't real, but that's largely because a lot of real photography is overproduced & ugly. I have yet to see anything that's aesthetically good.
And you see this in the other examples too. The teach the CLT piece was funny. It's and error-ridden mess that isn't even an explanation at all!
The best part is, someone looked at that and thought yes, that's right. Shows you clearly why programmers will be needed in the future. The LLM might be able to run circles around that person in math, but that person also has no idea when they're being bullshitted.
Glad to see it happen. I'd prefer it there was just one format rather than both jxl and avif, but at least it's the final nail in the coffin of webp which seems to have accomplished little but annoy people for not much gain.
The wider ecosystem support is still far from commonplace, but it is slowly changing. iOS 18 wouldn't work with .jxl in Photos, but 27 does. Similarly, quick look and preview works fine on MacOS 27, thumbnails show up properly etc. I also found no issues with .jxl on linux, and image editors are slowly adding support as well.
Plain old jpg will still be everywhere for the next decade, but I'm glad that we finally have superior options with no real downsides, and without all the patent bullshit to boot.
People are saying "Bad Support", but that's irrelevant without there being a popular place that supports it and has a LOT of it, and a bunch of other popular places that do not support them.
So yeah.
The REAL reason for WebP hate:
Google started converting image results to WebP automatically. Absolutely, officially diagnosed, mentally deficient move. Having every kid download a image of google and be unable to send it to discord made an ARMY of people who call it the "bad poopoo cringe format for s"
JXL is too new to draw any conclusions. But indeed, the market has spoken and it doesn't want avif. I tried adding avif support to a service myself, but the very slow encode speed made it unsuitable without paying for a larger server (or worse, a thumbnailner-as-a-service). Only a handful of enthusiasts (who chime in every such thread!) like avif.
Personally I fully expect jxl to succeed where avif failed and webp floundered.
It's frankly pointless. Pretty much nothing produces them natively, so it's usually a generation loss from JPEG, not enough compression gains without annoyingly visible loss of quality, still has format many format and functional limits so it solved nobody's problem at any time, pretty much.
And much (though not all) of its improved compression (before it damages the image too badly) was anyway matched by the better JPEG encoders which were released and improved over time. It sure benched very well against a 30+ year old JPEG encoder implementation though.
JXL has a pixel-accurate JPEG-input mode with no generation loss which already beats WebP, a native mode which is way way better even, it addresses many more use cases and is not Google's pet project (saying this while thanking them for all the foundational work from which modern image/video codecs benefit - JXL itself first and foremost!) so it DOES have better than a snowball's chance in hell of becoming the defacto standard of the next 10-30 years and we'll soon see it in silicon.
Actually (:nerdface:) byte-accurate / DCT-coefficient-accurate, but using a JXL decoder to decode such a JXL to pixels directly (instead of roundtripping through JPEG) can produce better quality than a typical JPEG decoder, due to the use of higher precision in decoding.
"webp" sounds like it's specifically targeted to the web. If I want to save an image in a certain file format, I am not going to chose the one that sounds like it's meant for deployment in web browsers, with potentially low quality for faster transfer.
While that might not be what WebP does, it is what it implies to regular users.
Well while you're here, there definitely are some issues with animated image support. Just earlier this week I attempted to use an animated .avif as my avatar image. It uploaded fine, the preview showed it fine, but the actual applied image was static again. Not sure if there's some issues with in the pipeline to handle those.
Surely support is even worse for both .jxl and .avif? If the annoying thing about .webp was lack of support, I don't think .jxl nor .avif solve any of that - just the opposite.
One thing to note is that even when WebP did eventually get support in many places, it often look a very long time because there just wasn’t that much motive or interest to support it. So for years it was the weird format you’d accidentally download from the web and have little luck getting to work off-line.
The non-exhaustive table linked below shows how long it took for different programs to support different image formats since their introduction. Outside of Chrome, Chrome in disguise, and whatever has become of Firefox, adoption of JPEG XL has been far more enthusiastic than it ever has for WebP—or AVIF, for that matter. So we are very unlikely to have a long period when JPEG XL only works on the web and nowhere else.
Right, due to slim support outside of browsers, webp files kinda turned into lumps of coal when downloaded, going from the image you wanted to a chunk of data that nothing but the browser knew what to do with. It was like the file format version of those fake-transparent PNGs with checkerboard background.
I wish gambling was outlawed entirely. Let bookies run it in some crack dens, I don't care about that, but every single gambling ad I see in the open is infuriating for me like no other ads. It's been seeping into sports commentary and even news on TV, "oh I don't think we have a full picture of the situation Kalshi has it at 60%", die in a fire. It's comparable to scammers preying on grandma.
No, I don't care about the "but liberty!" arguments.
For AI? Prompt comes in, tokens come out. The value of the 'ecosystem' is roughly zero to me. Switching costs between models are essentially zero. The UI and UX is the same whichever LLM runs underneath.
If you do use llm some, then the $20 tier makes sense, but that's not where the money is. If you use it a lot then a $100 sub probably makes sense. But if you're a BigCo that heavily uses it and pay API prices and suddenly you pay $200 million a month, well, maybe it's worth buying a bunch of hardware and stick it in a basement somewhere, and you won't upload all your secrets to your future competitor to boot. If US labs didn't care about the competition from Chinese models they wouldn't lobby so hard for banning the evil, unaligned, super dangerous communist models that want to steal our freedoms and make our gas expensive.
Ultra-cheap models like Deepseek are also very nice if you need to burn billions and billions of tokens.
reply